Light Isotope Nucleosynthesis with JAX (LINX) is a fast, differentiable, extensible code that can be used to predict primordial light element abundances during Big Bang Nucleosynthesis (BBN). LINX is written in python using JAX, and is readable, accessible, and well-documented. Full documentation is available at https://linx.readthedocs.io/en/latest/.
It is strongly recommended that you create a clean conda environment with the latest python. All of the requirements are included in requirements.txt. In summary, run
conda create --name linx
conda activate linx
pip install -r requirements.txt
(Include path/to/requirements.txt if you're in a different directory.) Verify your JAX installation by opening up python in terminal and attempting to import jax.
If you'd like to install a minimal set of packages just to run LINX without the SVI examples, you can run
conda create --name linx
conda activate linx
conda install --yes scipy matplotlib numpy jupyter
pip install diffrax==0.4.1 jax==0.4.28 jaxlib==0.4.28
(pip or conda can be used for scipy, matplotlib, numpy, and jupyter.)
If you have trouble installing JAX, we have a troubleshooting document with some tips.
Some analysis files in this repository depend on the user having installed CLASS and clik. Installation instructions are available for each of these codes: CLASS, clik. Note that some users have reported issues using clik with python 3.12--we will not maintain this note to check if this compatibility issue is resolved, but if you have difficulty installing clik you might try to install with python 3.11.
Modules of the code are contained in the linx directory, along with the various sets of BBN reaction rates that ship with LINX.
example_notebooks contains a set of pedagogical Jupyter noteboooks for users familiarizing themselves with LINX.
- background_evolution explores the thermodynamic calculation performed in LINX to determine quantities like energy densities and Hubble. It includes a comparison of the LINX Neff results with other results in the literature.
- NuisanceParametersImpact explores the impact of the uncertainties of BBN rates on final predicted abundances. The ability to sample these nuisance parameters is a key feature of LINX.
- weak_rates illustrates the impact of different settings for computing the rate of proton-neutron interconversion on the prediction for the primordial helium-4 abundance.
- Schramm includes an example of a Schramm plot, or the primordial abundances as a function of the baryon density. It includes examples with the truncated and full networks, as well as with nuclear rate uncertainties incorporated into the results.
- nuclear_evolution examines the evolution of primordial abundances as a function of time during BBN. It also includes a cool movie!
- inference_omega_b provides a toy example of parameter inference with LINX, with a profile likelihood test.
- scan_over_Omega_b_tau_n includes a scan over two parameters, both the baryon density and the neutron lifetime.
scripts contains a suite of scripts for testing and using LINX.
- test scripts: The
test_SBBNscripts test "standard BBN", or BBN in LCDM cosmology, using a truncated reaction network. This reaction network is the minimal network required for accurate deuterium and helium-4 predictions, and these two test scripts compare the LINX results for this network against results from PRIMAT and PRyMordial. Thetest_full_networkscripts are similar, but test the full network instead of the truncated network. - Analysis scripts: A set of python scripts and batch scripts used for the analyses in our analysis paper (coming soon!). These are the scripts that begin with
BBN_only...orCMB_BBN.... The CMB_BBN script requires the user to have installed CLASS and clik. run_numpyro.py: a script used to perform a Stochastic Variational Inference (SVI) analysis with LINX and CosmoPower in our formalism paper (coming soon!). Running this scripts requires the user to have installed CosmoPower.
This directory also contains two subdirectories: samples, which includes the data in our analysis paper (coming soon!), and related_notebooks, which contain jupyter notebooks that outline how to analyze the data generated by these scripts and stored in samples (requires the user to have corner installed). Important: The samples are somewhat large, taking up ~72MB of space, so they are not included in our main branch. If you'd like to see the samples, pull the samples branch instead.
Finally, pytest contains brief pytest scripts for continuous checking of this repository. You can run them yourself by navigating into the directory and running pytest . (after pip installing pytest).
Please feel free to open up an issue if you notice something is amiss in LINX.
If you use LINX, we suggest you cite both LINX and the other public codes that provide a foundation for LINX: nudec_BSM (https://arxiv.org/abs/1812.05605, https://arxiv.org/pdf/2001.04466), PRIMAT (http://www2.iap.fr/users/pitrou/primat.htm), and PRyMordial (https://arxiv.org/abs/2307.07061).
The suggested language for citation is "LINX [linx], which uses methods and tables from Refs. [nudecBSM1,nudecBSM2,PRIMAT,PRyM]". All of these citations are included below.
@misc{linx,
title={{LINX}: A Fast, Differentiable, and Extensible Big Bang Nucleosynthesis Package},
author={Cara Giovanetti and Mariangela Lisanti and Hongwan Liu and Siddharth Mishra-Sharma and Joshua T. Ruderman},
year={2024},
eprint={2408.14538},
archivePrefix={arXiv},
primaryClass={astro-ph.CO},
url={https://arxiv.org/abs/2408.14538},
}
@article{nudecBSM1,
title={Neutrino decoupling beyond the {S}tandard {M}odel: {CMB} constraints on the Dark Matter mass with a fast and precise {N}eff evaluation},
volume={2019},
ISSN={1475-7516},
url={http://dx.doi.org/10.1088/1475-7516/2019/02/007},
DOI={10.1088/1475-7516/2019/02/007},
number={02},
journal={Journal of Cosmology and Astroparticle Physics},
publisher={IOP Publishing},
author={Escudero, Miguel},
year={2019},
month=feb, pages={007–007} }
@article{nudecBSM2,
title={Precision early universe thermodynamics made simple:{N}eff and neutrino decoupling in the {S}tandard {M}odel and beyond},
volume={2020},
ISSN={1475-7516},
url={http://dx.doi.org/10.1088/1475-7516/2020/05/048},
DOI={10.1088/1475-7516/2020/05/048},
number={05},
journal={Journal of Cosmology and Astroparticle Physics},
publisher={IOP Publishing},
author={Abenza, Miguel Escudero},
year={2020},
month=may, pages={048–048} }
@article{PRIMAT,
title={Precision big bang nucleosynthesis with improved {H}elium-4 predictions},
volume={754},
ISSN={0370-1573},
url={http://dx.doi.org/10.1016/j.physrep.2018.04.005},
DOI={10.1016/j.physrep.2018.04.005},
journal={Physics Reports},
publisher={Elsevier BV},
author={Pitrou, Cyril and Coc, Alain and Uzan, Jean-Philippe and Vangioni, Elisabeth},
year={2018},
month=sep, pages={1–66} }
@article{PRyM,
author = "Burns, Anne-Katherine and Tait, Tim M. P. and Valli, Mauro",
title = "{PRyMordial: the first three minutes, within and beyond the Standard Model}",
eprint = "2307.07061",
archivePrefix = "arXiv",
primaryClass = "hep-ph",
reportNumber = "UCI-HEP-TR-2023-07, YITP-SB-2023-16",
doi = "10.1140/epjc/s10052-024-12442-0",
journal = "Eur. Phys. J. C",
volume = "84",
number = "1", pages = "86",
year = "2024" }
A detailed reference outlining the structure and philosophy of LINX and including examples of how to use it with gradient-assisted methods is available at https://arxiv.org/abs/2408.14538.
An example of an analysis that can be performed with LINX, where LINX is used with CLASS to perform an analysis with a joint CMB and BBN likelihood, is available at https://arxiv.org/abs/2408.14531.